IÉSEG Insights » Resource Center » Big Data & AI » Are we more likely to believe AI when it is wrong?

Are we more likely to believe AI when it is wrong?

08/25/2026

3 min

Artificial intelligence (AI) is everywhere in the workplace, supporting workers in a wide range of tasks such as medical diagnosis, recruitment and product design. This reflects the idea that humans and AI are more effective when working together than alone. But is this always the case? In this video, Professor Domenico DI PRISCO, shares the findings of his research on the way we react to AI recommendations.

People respond to AI recommendations in very different ways. Some follow them blindly, whilst others reject them outright. So, what drives their responses?

“Our research suggests that reactions depend on the mental models or frames that help us interpret situations, guide decisions and shape our expectations about plausible outcomes,” notes the expert.

Expectations & AI recommendations

However, people’s expectations are not always met. When this occurs and mental frames are disrupted, how do people respond to AI recommendations?

“Across three experiments, my co-author, Silvia Dello Russo, and I, found a clear pattern. When expectations are met, people tend to dismiss contradictory AI suggestions, often assuming the AI is malfunctioning.” But when unexpected events challenge our mental models, something different happens.

“People stop trying to solve the original problem on their own. Instead, they focus on making sense of the AI’s suggestion. People assume the AI must be correct and believe the issue lies in their own inability to understand it. And if they cannot find an explanation, they are more likely to blindly accept the suggestion.”

Surprisingly, this blind reliance is even stronger when AI outputs are not only difficult to interpret but clearly wrong or absurd.

“In these cases, using AI to enhance our efforts or decisions can be time-consuming and even risky. People may read meaning into incorrect outputs, leading to major failures.”

Given that both humans and AI struggle to make correct decisions when facing rare circumstances, their findings also suggest that people are more likely to be influenced by AI precisely when its predictions are also more likely to be incorrect.

“Collaboration between AI and humans is, of course, valuable. However, our work highlights important challenges and some of the risks that must be considered to ensure the responsible use of AI in organizations and society.”

Find out more in the research article

Sensemaking and AI: Unraveling individuals’ reactions to the black box in a three-study investigation (Technological Forecasting and Social Change, 2026) Domenico di Prisco (IÉSEG) & Silvia Dello Russo (Luiss University & Luiss Business School)